[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Bug default/19427] Intern the strings used in Libabigail
https://sourceware.org/bugzilla/show_bug.cgi?id=19427
--- Comment #4 from dodji at seketeli dot org ---
> Not that this is any of my business :-)
It is, otherwise, this would not be an open project ;-)
> The 2% speed difference is insignificant. No-one will care.
>
> The memory savings are nice, but not *that* huge (around 7%). I'd be inclined
> to make this kind of change only if the impact on code complexity and
> maintainability is low.
Actually, someone brought to my attention a test case where we have 30%
speed increase ;-)
This is on the nss package, from mozilla. That is why I merged it in.
In other words, it makes the library scale a lot more on packages with
deep aggregate type hierarchies where each aggregate type has a lot of data
members, and where sub-types reference each other. If you also have a
lot of incomplete types in the mix, then you end up doing a *lot* of
type name comparisons. And this is where interned strings help.
> I've worked with code bases that create their own string implementation in the
> past and, usually, it creates more problems than it solves. Often, the problem
> ends up being that the replacement cannot be easily exchanged with std::string
> and vice-versa, and then ends up costing more than it saves. So, whenever I see
> a string replacement class, I tend to look twice.
Right, I totally agree.
This is why I tried very hard to make abigail::interned_string to be
just a pointer to std::string, in essence. It's not a new string. The
idea is just that comparing two instances or interned_string amounts to
comparing the values of the pointers to std::string, as there is only
one copy of each distinct std::string in the entire environment now.
> I don't know whether you can use boost in your code.
As this thing should work on systems that are not updated that much, for
instance, el6, we don't use boost, no.
--
You are receiving this mail because:
You are on the CC list for the bug.